ARTICLE

DOI: 10.1038/s41467-017-00197-0 OPEN Spiraling pathways of global deep waters to the surface of the Southern Ocean

Veronica Tamsitt 1, Henri F. Drake 2,6, Adele K. Morrison2,7, Lynne D. Talley 1, Carolina O. Dufour2, Alison R. Gray 2, Stephen M. Griffies3, Matthew R. Mazloff1, Jorge L. Sarmiento2, Jinbo Wang4 & Wilbert Weijer5

Upwelling of global deep waters to the sea surface in the Southern Ocean closes the global overturning circulation and is fundamentally important for oceanic uptake of carbon and heat, nutrient resupply for sustaining oceanic biological production, and the melt rate of ice shelves. However, the exact pathways and role of topography in Southern Ocean upwelling remain largely unknown. Here we show detailed upwelling pathways in three dimensions, using hydrographic observations and particle tracking in high-resolution models. The analysis reveals that the northern-sourced deep waters enter the Antarctic Circumpolar Current via southward flow along the boundaries of the three ocean basins, before spiraling southeastward and upward through the Antarctic Circumpolar Current. Upwelling is greatly enhanced at five major topographic features, associated with vigorous mesoscale eddy activity. Deep water reaches the upper ocean predominantly south of the Antarctic Circumpolar Current, with a spatially nonuniform distribution. The timescale for half of the deep water to upwell from 30° S to the mixed layer is ~60–90 years.

1 Scripps Institution of Oceanography, La Jolla, CA 92093, USA. 2 Princeton University, Princeton, NJ 08544, USA. 3 Geophysical Fluid Dynamics Laboratory, Princeton, NJ 08540, USA. 4 Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109, USA. 5 Los Alamos National Laboratory, Los Alamos, NM 87545, USA. 6Present address: Massachusetts Institute of Technology and Woods Hole Oceanographic Institution Joint Program in Oceanography, Cambridge, MA, USA. 7Present address: Australian National University, Canberra, ACT 2602, Australia. Correspondence and requests for materials should be addressed to V.T. (email: [email protected])

NATURE COMMUNICATIONS | (2017) 8:172 | DOI: 10.1038/s41467-017-00197-0 | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00197-0

he global overturning circulation moves waters around the a Australia Tworld’s oceans, connecting surface and deep waters through two interlinked overturning cells, one with sinking in the far northern North Atlantic and adjacent Nordic Seas and the other with sinking along the Antarctic coastline1, 2. These pro- cesses are well documented, with the northern sites well mapped and the southern sites, in coastal polynyas, increasingly so3.In contrast, the specific locations where these waters return back to the sea surface to complete the circuit are poorly known. Obser- vations suggest that as much as 80% of the World Ocean deep water returns to the surface in the Southern Ocean with the S. Africa remainder reaching the sea surface through upwelling to the 2, 4 0 S. America thermocline in low latitudes . The vigor of the Southern Ocean –3 return limb derives from the dynamics associated with the Neutral density surface 28.05 kg m existence of an open circumpolar pathway around Antarctica in b Drake Passage latitudes5. Dense deep water is drawn upward 500 Australia along steeply tilted isopycnals (surfaces of constant density), dri- ven by divergence of wind-driven Ekman transport and surface buoyancy forcing, enabling the return of deep water to the surface 1000 with minimal diapycnal mixing6, 7. In the upper overturning cell, this upwelled water is transported northward via wind forcing and 1500 becomes lighter mode and intermediate waters. Below this, in the lower cell, the upwelled water is transformed into abyssal Ant- Depth (m) arctic bottom water (AABW) that sinks, moves northward, and is 2000 S. Africa then converted to deep waters through diabatic mixing above the – seafloor8 10. The warm, upwelled water that nears the ice shelves 11 of West Antarctica is recognized as a major factor in the high 2500 S. America rate of ice shelf basal melt;12 variability in upwelling is therefore Atlantic particle pathways one likely contributor to the accelerated melt rate documented in c Australia this region13, with long-term consequences for sea level rise. 3000 This major Southern Ocean return limb of the global over- turning circulation is usually described in a two-dimensional sense (latitude-depth space), drawing on its parallel with the strongly zonally symmetric atmospheric dynamics. Mesoscale eddies have long been recognized as fundamental to the zonally averaged view of the Antarctic Circumpolar Current (ACC), arising due to baroclinic instability associated with the sharply sloped isopycnals. In the upper ocean, southward eddy-induced S. Africa transport directly opposes the northward Ekman transport, lim- iting the residual overturning magnitude and reducing the sen- sitivity of the overturning to strengthening westerly winds14, 15. S. America Example Atlantic particle trajectories Beneath the surface layer, eddies are the primary mechanism for the southward transport of deep water across the ACC fronts16,in Fig. 1 The three dimensional upward spiral of North Atlantic Deep Water the latitude and depth range that is unblocked by continental through the Southern Ocean. a Observed warm water (>1.6 °C) on the boundaries or topographic ridges (“Drake Passage effect”)1, 17. 28.05 kg m−3 neutral density surface from hydrographic observations27, However, recent studies have demonstrated strong zonal varia- south of 40° S, colored by depth (m). The isoneutral surface is masked in tions in the Southern Ocean circulation, emphasizing the regions with potential temperature below 1.6 °C. 1/4° ocean bathymetry70 importance of taking into account the three-dimensionality of the is shown in gray. b Modeled (CM2.6) particle pathways from the – circulation2, 16, 18 20. Atlantic Ocean, with particles released in the depth range 1000–3500 m The southeastward pathway that the deep waters follow, entering along 30° S. Colored boxes mark 1° latitude × 1° longitude × 100 m depth grid from the basins lying to the north and then traveling around boxes visited by >3.5% of the total upwelling particle-transport from Antarctica until reaching the continental margin, is an aspect of the release at 30° S to the mixed layer. Boxes are colored by depth, similar to a. Southern Ocean circulation that is familiar from maps of the surface c Two example upwelling particle trajectories from CM2.6, one originating circulation. However, this circulation is rarely explored for its from the western Atlantic and the other from the eastern Atlantic. interaction with the upwelling of the deep waters along this path, Trajectories are colored by depth as in a and b, blue spheres show the and for the specific locations where enhanced upwelling occurs. The particle release locations and red spheres show the location where the ACC spirals southeastward from its northernmost latitude just east particles reach the mixed layer. Three-dimensional maps were produced of South America to its entry into Drake Passage from the Pacific, using Python and Mayavi71 nearly 1700 km farther south21. The southward shift is consistent with a vorticity balance in which mean advection of planetary The time scale for deep waters to reach the sea surface from vorticity by the ACC balances vorticity generation by wind each of the northern basins is important, both for setting the stress curl (i.e., Sverdrup balance). Previous work has noted the temporal response in the Southern Ocean to major changes in existence of a spiral structure in the Southern Ocean upwelling22. northern deep water formation rates23 and for its control on However, to date the detailed geographic distribution and biogeochemical processes that affect climate24, 25. Relatively mechanisms for the upwelling along this ACC path have been carbon-poor North Atlantic Deep Water (NADW) mingles with largely unexplored. much older, carbon-rich Indian and Pacific Deep Waters (IDW

2 NATURE COMMUNICATIONS | (2017) 8:172 | DOI: 10.1038/s41467-017-00197-0 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00197-0 ARTICLE

2 and PDW) and all rise to the surface . The relative amounts and a Pacific Atlantic Indian time scales of these different northern deep water components 0 impact the near-surface, upwelled ocean carbon and nutrient –10 content and the heat supply to the Antarctic margins. We document here the three-dimensionality of upwelling from –20 CESM CM2.6 Transport (Sv) Transport –30 the deep ocean interior to the surface of the Southern Ocean with SOSE observations and three independent, state-of-the-art, eddying 180° W 120° W 60° W600° E ° E 120° E ocean and climate models. Our analysis reveals the locations 0 where the deep waters are most strongly shifted upwards and b fi where they reach the sea surface. We nd that upwelling along the 1000 southeastward spiral is not uniform. Where the ACC encounters major topographic features, flow-topography interactions create 2000 localized energetic eddy “hotspots”26, which drive enhanced cross- frontal exchange16, 19. Here we show for the first time that deep Depth (m) water upwelling is also strongly enhanced at these hotspots. We 3000 also give a first estimate of the time scales of this upwelling and CESM 4000 relative contributions of deep waters from the Atlantic, Indian, 180° W 120° W 60° W600° E ° E 120° E and Pacific to upwelled waters in the Southern Ocean. 0 2 c ) –1 1000 1

Results Sv m –2 Three-dimensional deep water spiral. The broad three- 2000 0 dimensional pathway of upwelling in the Southern Ocean is 27 Depth (m) illustrated using observed properties along a surface repre- 3000 −3 –1 senting NADW (the neutral density surface 28.05 kg m ; Fig. 1a, (x10 Transport CM2.6 Supplementary Fig. 1). The relatively warm, saline NADW, 4000 represented in Figure 1a by waters warmer than 1.6 °C, enters the 180° W 120° W 60° W600° E ° E 120° E –2 Southern Ocean from the deep Atlantic (2800 m depth) and 0 spirals southeastward and upward through the ACC. Waters d warmer than 1 °C on this neutral density surface approach the 1000 Antarctic continental shelf (500 m depth) along the West Ant- arctic Peninsula and Amundsen Shelf south of 60° S, where 2000 incursions of upwelled, warm, northern-sourced deep waters have 28 Depth (m) been implicated in the accelerated melting of ice shelves . 3000 Associated maps show the separate entrances of high-nutrient/ SOSE low-oxygen IDW and PDW into the southeastward spiral (Sup- 4000 plementary Fig. 1 and Supplementary Note 1)27. The spiraling 180° W 120° W60° W600° E ° E 120° E paths of NADW/IDW/PDW properties mostly follow the ACC Fig. 2 Model comparison of volume transports at 30° S. a Shows the fronts, and, upon close inspection, appear to cross fronts down- volume transport in Sverdrups integrated over the depth range 1000–3500 stream of major topographic features (Supplementary Fig. 1). m, b, c and d show the Eastward cumulative integrals of the time average More detailed geographic description and timescales of meridional transport in Sv m−1 at 30° S in CESM, CM2.6, and SOSE, upwelling are difficult with the sparse Southern Ocean hydro- respectively. The transports in b–d are normalized by the model vertical graphic data sets. We therefore use a Lagrangian modeling grid thicknesses approach to quantify Southern Ocean upwelling and explore mechanisms controlling its pathways. We track virtual particles models). Additionally, the model (Fig. 1b) also shows the preferred and their associated volume transports (particle transport) from boundary current pathways from 30° S and the near-surface the deep ocean interior (1000–3500 m layer) at 30° S until they continuation of the NADW pathway along the Antarctic coast, reach the mixed layer in three independent eddying models: the which is unclear in the NADW temperature maximum due to Community Earth System Model (CESM), the Geophysical Fluid mixing with colder surrounding waters before reaching Antarctica. Dynamics Laboratory’s Climate Model version 2.6 (CM2.6), and A comparison of the time-mean volume meridional transport the Southern Ocean State Estimate (SOSE; See Methods section at 30° S in CESM, CM2.6, and SOSE shows reasonable agreement for model and particle tracking details). We note that while the in the magnitude and spatial structure of volume transport 1000–3500 m depth range spans a broad range of deep water (Fig. 2). The vertically integrated southward volume transports in densities, the focus here is on interior upwelling away from the 1000–3500 m depth range agree closely in the Pacific, with the bottom boundary layer processes, rather than the upwelling of largest differences in the western Atlantic and western Indian AABW from the abyssal ocean. Ocean. The total Eulerian southward transport across 30° S Modeled particles from the deep Atlantic preferentially spiral between 1000 and 3500 m is 28.8, 22.7, and 32.9 Sv in the CESM, southeastward and upwards through the ACC (Fig. 1b, c and CM2.6, and SOSE, respectively. These southward transports are Supplementary Movie 1 using CM2.6; CESM and SOSE results are slightly larger than the net transport in the southward limb of the qualitatively similar). Similar spirals are also clear for modeled zonally averaged overturning streamfunction (Fig. 2; 24.4, 21.1, particles released in the Pacific and Indian Oceans (Supplementary and 29.0 Sv in the CESM, CM2.6, and SOSE, respectively). Fig. 2, Supplementary Movies 2 and 3, and Supplementary Note 2). By comparison, the total Lagrangian upwelling particle The modeled Atlantic spiral (Fig. 1b) strongly resembles the transport reaching the mixed layer south of 30° S is 13.2, 11.6, observed pathway of the warm, saline NADW (Fig. 1a), although a and 21.3 Sv in CESM, CM2.6, and SOSE, respectively. For all different diagnostic is used (temperature on an isopycnal for the models, the Lagrangian transports are less than the Eulerian and observations and probability of passing through a grid box for the overturning streamfunction transports. The two are not expected

NATURE COMMUNICATIONS | (2017) 8:172 | DOI: 10.1038/s41467-017-00197-0 | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00197-0

abc 34.0 34.0 34.0 30 CESM CM2.6 SOSE ) ) ) 34.5 34.5

–3 34.5 –3 20 –3

35.0 35.0 35.0 10 35.5 35.5 35.5 0 36.0 36.0 36.0 –10 Transport (Sv) Transport 36.5 36.5 36.5 –20 Potential density (kg m Potential Potential density (kg m Potential Potential density (kg m Potential 37.0 37.0 37.0 –30 S S S S S S S S S S S S S S S ° ° ° ° ° ° ° ° ° ° ° ° ° ° ° 70 60 50 40 30 70 60 50 40 30 70 60 50 40 30 Fig. 3 Model comparison of Southern Ocean zonally averaged circulation. Meridional overturning streamfunction (Sverdrups) in a CESM, b CM2.6, and c SOSE. Solid and dashed contours represent positive and negative transport, respectively, with an interval of 2.5 Sv to agree in this case, and this is largely due to our definition of broadly below both the East Australian Current in the Tasman Lagrangian transport, whereby we only select particle trajectories Sea and East Auckland Current around New Zealand and out into that reach the mixed layer. In the overturning streamfunction, the deep Pacific following topography. Another deep boundary there is a portion of the southward upwelling limb that is pathway in the mid-Indian Ocean follows topographic ridges, entrained into either intermediate or abyssal waters in the interior especially the Southwest Indian Ridge32. without ever reaching the mixed layer. Additionally, there is The eastern pathways in each ocean basin are less documented likely a small fraction of Lagrangian particle-transport that takes than the DWBCs. Part of the NADW leaves the Atlantic just west longer than 200 years to upwell and thus is not captured in of South Africa, having crossed the South Atlantic at mid-latitude, our total transport. NADW dominates the total upwelling consistent with both observations and models33, 34. This pathway particle-transport in all three models (51% in CM2.6 and CESM, is hypothesized to be driven by southward eddy thickness fluxes 41% in SOSE), with the remaining transport split almost equally imposed by the northwestward movement of shallow Agulhas between the IDW and PDW. rings30. While the current has been identified in observations at The time taken for particles to travel from 30° S to the mixed 115°E by its eastward transport and low-oxygen content35, layer is in the range of decades to more than a century, with peak characteristic of IDW36, its global impact has not been upwelling occurring at 41, 28, and 81 years after release in CESM, appreciated and its physical cause has not been shown. We CM2.6, and SOSE, respectively (Fig. 3a, Supplementary Table 1). hypothesize that the eastern Atlantic eddy thickness flux We note that these transit times are considerably faster than mechanism30 may also operate in the Indian, driven by eddy a previous estimate of 140 years from a relatively coarse transport south and west of Tasmania and flowing along the resolution (non-eddying) model29. We hypothesize from this southern coast of Australia37. In the eastern Pacific, a broad previous study and our results that upwelling timescales are meandering pathway carries PDW southward, as identified in resolution dependent, which would explain the slower upwelling hydrographic observations38, 39. An inverse model of the in the 1/6° SOSE compared to the 1/10° CESM and CM2.6. In Southeast Pacific circulation indicates that eddies likely play an CM2.6 and CESM, the median upwelling time for particle important role in this pathway, but more work is needed to transport originating in the Indian Ocean is slightly longer than understand the underlying dynamics39. the Atlantic and Pacific, while in SOSE, particle transport from Although there is good agreement on the location of pathways the Pacific takes substantially longer to upwell than from the in the three models, there are differences in the relative strengths Indian and Atlantic (Fig. 4b–d). There is a distinct difference in of individual upwelling pathways. In particular, the contribution upwelling from the Indian in SOSE relative to CESM and CM2.6, of the Pacific to the total particle transport is relatively large in with large initial upwelling in the first 25 years (Fig. 4c, green SOSE, and the strength of the eastern Indian and Pacific pathways line). This may arise from the relatively large particle transport varies significantly across the models. These differences are likely carried along the western boundary of the Indian Ocean by the attributable to differences in meridional transport at 30° S in each Agulhas current in SOSE, which leads to rapid coastal upwelling model (Fig. 1) or differences in model spatial and temporal from the depths in the shallower part of the 1000–3500 m range. resolution (see Methods section). Thus, we focus on the features The three-dimensional upwelling picture (Fig. 1b) is quantified that are common to all three models. for particle trajectories from all ocean basins in a two- dimensional view (Fig. 5), revealing the horizontal pathways of upwelling and their relative strengths. Particle transport originat- Topographic upwelling hotspots. Figure 5 shows the spatial ing in the Atlantic, Indian, and Pacific Oceans at 30° S flows distribution of particles at their final crossing of depth surfaces southward before merging into the ACC. From here, up to 20% of while upwelling. Upwelling in the ocean interior within the the total particle transport, depending on the model and basin of southeastward spiral is concentrated at the five major topographic origin, move into parts of the Ross and Weddell Gyres and along features crossed by the ACC (shown in Fig. 6a, b for CM2.6 the Antarctic coast. The pathways in Fig. 5 are remarkably and Supplementary Figs. 6 and 7 for CESM and SOSE). These insensitive to minor variations in the Lagrangian method hotspots dominate the total upwelling across depth surfaces, with (Supplementary Figs. 3–5 and Supplementary Note 3). There >55% of the total particle-transport upwelling across the 1000 m are two distinct types of pathways into the ACC: via deep western depth surface occurring in these five topographic hotspots in all boundary currents (DWBCs) along continents or topographic three models, which span only 25% of the total zonal extent of the ridges, and along eastern pathways whose dynamics may be eddy- Southern Ocean (shaded in gray in Fig. 6a). These hotspots occur driven30. DWBCs are the shortest and fastest routes and have within the ACC boundaries, so that most of the upwelling been previously identified in Lagrangian experiments31. The across 1000 m occurs within the ACC latitude range, between DWBCs carry deep water beneath the Brazil Current in the 40 and 60° S (Fig. 6c). We note that there is also enhanced Atlantic, beneath the Agulhas Return Current in the Indian, and upwelling north of the ACC in the southward flowing western

4 NATURE COMMUNICATIONS | (2017) 8:172 | DOI: 10.1038/s41467-017-00197-0 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00197-0 ARTICLE

a 1.4 b 1.4 CESM 1.2 All CM2.6 1.2 Atlantic 1.0 SOSE 1.0 0.8 0.8 0.6 0.6 0.4 0.4 % particle-transport

0.2 % particle-transport 0.2 0.0 0.0 0 50 100 150 200 0 50 100 150 200 Time of Lagrangian integration (years) Time of Lagrangian integration (years)

c 1.4 d 1.4 1.2 Indian1.2 Pacific 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 % particle-transport % particle-transport 0.2 0.2 0.0 0.0 0 50 100 150 200 0 50 100 150 200 Time of Lagrangian integration (years) Time of Lagrangian integration (years)

Fig. 4 Particle upwelling timescales. Transit time distribution for particle-transport from 30° S to the mixed layer in the three models for particles originating in a all basins, b the Atlantic, c the Indian, and d the Pacific boundary currents, but we focus our attention on the mechanism enhanced upwelling into the surface ocean at topographic for upwelling hotspots within the ACC. features42. For example, upwelling across 200 m is enhanced in The strongly localized distribution of upwelling at 1000 m all three models at the Kerguelen Plateau, Macquarie Ridge, and differs from the uniform upwelling expected from wind stress Pacific–Antarctic Ridge, although there are substantial differences curl over the Southern Ocean. The hotspots of upwelling in the relative importance of these hotspots at 200 m between the within the ACC at 1000 m occur in regions of high eddy models (Fig. 6f and Supplementary Figs. 6 and 7). These kinetic energy (EKE, see Methods section for definition) associated differences in particle transport at the 200 m depth surface with topography (blue contours in Fig. 6a, b, Supplementary compared to 1000 m indicates that differences in upper ocean Figs. 6 and 7, and Supplementary Note 4), where interactions processes between models impact the 200 m upwelling distribu- between the mean flow and topography enhance eddy tion, although lower spatial resolution could also contribute to the activity19, 40. Recent studies have shown preferential southward difference between SOSE and the two higher resolution models. transport of particles and tracers across ACC fronts in the upper A schematic of a representative Southern Ocean upwelling 1500 m at topographic hotspots16, 19. Our results show the central pathway along an isopycnal surface is shown in Figure 7. Deep role of these same topographic hotspots in raising particles toward waters move southward from 30° S along isopycnals that are at the surface as they follow the ACC path. The mean particle roughly constant depth, primarily in deep boundary currents, transport crossing 1000 m in all regions where EKE exceeds 75 until joining the ACC where they follow the meandering paths of − cm2 s 2 is an order of magnitude larger than the mean elsewhere, the ACC fronts (red pathway in Fig. 7). Eddy advection drives and there are statistically significant correlations between mean flow across the ACC fronts in the ocean interior (yellow arrows). EKE at 1000 m and particle transport crossing the 1000 m depth Within the ACC, isopycnals slope strongly upwards towards the surface within the ACC of 0.33, 0.65, and 0.56 in CESM, CM2.6, surface, and simultaneously thin towards the south (Fig. 7b). and SOSE, respectively (Pearsons correlation coefficient with p- Eddies act to reduce the meridional thickness gradients, hence value < 0.01). Within the ACC, EKE and upwelling at 1000 m are advecting water southward and upward along isopycnals. The not expected to align perfectly, because all upwelling hotspots are upwelling pathways indicate that, between topographic features, associated with elevated EKE, while not all regions with high EKE particles primarily follow mean ACC streamlines around also have enhanced upwelling. Only locations that lie along the Antarctica (Fig. 5 and schematically in Fig. 7). Where ACC three-dimensional deep water pathways (Figs. 1b and 5)atthe fronts encounter topographic features, baroclinicity increases; 1000 m depth surface will show enhanced upwelling. strong eddy fields then develop downstream of topography19, The upwelled water in the three models reaches the surface advecting water southwards and upwards along isopycnals. layer, represented by upwelling across 200 m (Fig. 6d–f and Therefore upwelling particles generally approach topographic Supplementary Figs. 6 and 7), mostly along the southern features along more northerly ACC fronts and at greater depths, boundary of the ACC and over broader spatial scales than the and exit downstream along more southerly ACC fronts and at interior upwelling hotspots. This upwelling coincides with a shallower depths (Fig. 7b). Thus, the three-dimensional spiral is a region of enhanced buoyancy gain by surface freshwater fluxes superposition of the large-scale southeastward path of the mean from melting sea ice41. The remaining upwelling transport ACC fronts from the Atlantic to the Pacific, and eddy-driven reaches the surface throughout the subpolar gyres and along the “steps” southward and upward across fronts at topographic Antarctic coastline, where it is exposed to buoyancy loss and may hotspots. This upwelling motion along particle trajectories can be contribute to the formation of ABW. Even at 200 m, the broad visualized as a spiral staircase. distribution of upwelling, which is consistent with the broad While we propose that along-isopycnal eddy transport is pattern of negative wind stress curl, contains some localized the dominant mechanism for upwelling at topographic hotspots enhancements associated with topographic hotspots (Fig. 6d). within the ACC, diapycnal mixing may also play a non-negligible This agrees with a previous Lagrangian analysis that found role in the upwelling of deep water at these hotspots. Observations

NATURE COMMUNICATIONS | (2017) 8:172 | DOI: 10.1038/s41467-017-00197-0 | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00197-0

Atlantic Indian Pacific

CM2.6 51% 26% 23%

SOSE 41% 30% 29%

CESM 52% 25% 24%

0 5 10 15 20 25 30 Particle-transport (%)

Fig. 5 Particle pathways from 30° S to the mixed layer. Maps of the percent of total basin upwelling particle transport visiting each 1° latitude × 1° longitude grid column at some time during the 200 year experiment from release at 30° S and before reaching the surface mixed layer for CM2.6, SOSE, and CESM. The percentages of particle transports originating in the Atlantic, Indian, and Pacific (release locations at 30° S marked in red) are shown separately, normalized by the total upwelling particle transport originating in each basin. The percentages in the center of each panel indicate the relative contribution of the Atlantic, Indian, and Pacific to the total upwelling particle transport in each model suggest that interior diapycnal mixing is an important component continent described here provide a framework for understanding – of the Southern Ocean overturning9, 43 45, particularly in the where relatively warm deep water is supplied to the Antarctic upper 1000 m and 1000–2000 m above the seafloor9. Additionally, continental shelf and the origin of changes in the heat content of it has been shown in Drake Passage that the strength of abyssal this water. Observations indicate that upwelling deep water pre- mixing is dependent on local eddy energy46. In this work, the ferentially reaches close to the Antarctic continent along the focus on upwelling at hotspots associated with enhanced eddy western Antarctic Peninsula (Fig. 1a), but further analysis of our activity is at mid-depths away from the surface or seafloor model results are needed to determine the regionality of supply of topography. In the mid-depth ocean, along-isopycnal processes deep water to the continental shelf in greater detail. are expected to dominate over diapycnal processes. An analysis of From our simulations, we find that the timescale for deep water the extent to which the interior upwelling pathways are adiabatic, in the 1000–3500 m depth range to travel from 30° S to the surface and quantification of the diapycnal density change along mixed layer is of the order of multiple decades to a century Lagrangian trajectories at the upwelling hotspots is outside the (Fig. 4). This upwelling timescale has implications for the time scope of this study and is the subject of ongoing work. taken for changes in the deep ocean to be relayed to the surface of the Southern Ocean where they can influence the atmosphere. For instance, the peak upwelling timescale (mode) from the three Discussion models for deep water to travel from 30° S in the Atlantic Ocean From our results, we propose a new paradigm for the upwelling to the surface of the Southern Ocean ranges from 28 to 81 years. branch of the Southern Ocean overturning circulation that Chloroflourocarbon-based estimates of the timescale for water consists of a three-dimensional spiral, with most of the subsurface from deep water formation sites in the North Atlantic to first upwelling concentrated at the five major topographic features reach 20° S are on the order of 30 years47. This suggests a com- encountered by the ACC (Fig. 6): the Southwest Indian Ridge, bined advective timescale from the northern North Atlantic to the Kerguelen Plateau, Macquarie Ridge, Pacific–Antarctic Ridge, Southern Ocean surface on the order of a century. This estimate is and Drake Passage. The spatial structure of upwelling and comparable to the time lag between abrupt climate changes in mechanisms highlighted in this study have important implications the Northern Hemisphere and Antarctica of 218 ± 92 years and for climate. Upwelling deep water along the Antarctic continental 208 ± 96 years for warm and cold events, respectively, estimated shelf has driven an observed acceleration in basal ice shelf melt in from ice core records23, which are likely propagated from the recent decades28. The three-dimensional pathways carrying deep Northern Hemisphere to Antarctica via the ocean. Additionally, water from the Atlantic, Indian, and Pacific to the Antarctic our estimates of Lagrangian particle transport show that NADW

6 NATURE COMMUNICATIONS | (2017) 8:172 | DOI: 10.1038/s41467-017-00197-0 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00197-0 ARTICLE

1000 m meridional integral EKE (cm2 s–2) 2 aSWIR KP MR PAR DP 80 CESM 60 CM2.6 1 40 SOSE 20 CM2.6 EKE 0 0 % particle-transport 1000 m zonal integral 30° S b c

40° S

50° S

60° S

70° S CM2.6 1000 m 60° E 120° E 180° E 120° W60° W 0246810 % particle-transport 400 m meridional integral 2 d

1

0 % particle-transport

200 m zonal integral 30° S e f

40° S

50° S

60° S

70° S CM2.6 200 m 60° E 120° E 180° E 120° W60° W 0 246810 % particle-transport 0.0 0.1 0.2 % particle-transport

Fig. 6 Upwelling of particles across depth horizons. a Percent of total upwelling particle transport crossing 1000 m (1000 m is chosen because it is representative of upwelling in the interior and lies above major topographic features) as a function of longitude, integrated across all latitudes for all three models; the blue line shows the mean eddy kinetic energy (EKE) at 1000 m in CM2.6 averaged between 30° S and Antarctica at each longitude; gray shaded bars show the location of the five major topographic upwelling hotspots: the Southwest Indian Ridge (SWIR), Kerguelen Plateau (KP), Macquarie Ridge (MR), Pacific–Antarctic Ridge (PAR), and Drake Passage (DP). b Percent of particle transport crossing 1000 m in each 1° latitude × 1° longitude grid box between release at 30° S and the mixed layer in CM2.6. Blue contours indicate regions where the mean EKE at 1000 m in CM2.6 is higher than 75 cm2 s−2. c Percent of particle transport crossing 1000 m depth as a function of latitude, integrated across all longitudes for all three models. d Same as a for 200 m, without EKE contours, e same as b for 200 m without EKE contours, and f same as c for 200 m. In all panels, we select the location at which particles cross depth surfaces for the final time along their trajectories. Qualitatively similar results are obtained from selecting first-crossing locations. Black contours in b and e are the outermost closed contours through Drake Passage of mean sea surface height in CM2.6, representing the path of the Antarctic Circumpolar Current dominates the total upwelling. This suggests that changes in the fluxes might also present localized patterns in relation to these deep Atlantic may have a disproportionate impact on the deep upwelling hotspots, as suggested by the distribution of anthropogenic water properties that reach the surface of the Southern Ocean, and carbon uptake in an earlier iteration of SOSE51. Further work is thus have a greater influence on heat exchange with the atmo- needed to determine the correspondence between the distribution of sphere and cryosphere and on delivery of warm water to the upwelling into the surface ocean shown here and surface observa- Antarctic continental shelf48. tions,andtowhatextenttheseupwellingpatternsinfluence spatial Our result may have ramifications for the air-sea exchange of distributions of carbon flux. The significant differences between the carbon dioxide, as variability in tracer uptake in the Southern Ocean models in location of the deep water outcrops (Fig. 6d), in contrast is likely related to upwelling strength49, 50. The spatial patterns of with the strong agreement in the preferred locations of interior where deep water enriched in natural carbon but lacking in upwelling (Fig. 6a), emphasizes the importance of improving in situ anthropogenic carbon reaches the upper ocean (Fig. 6eandSup- observations of upwelling and carbon dioxide fluxes, which have high plementary Figs. 6 and 7) are highly localized, suggesting that carbon uncertainty due to sparse observations and large interannual

NATURE COMMUNICATIONS | (2017) 8:172 | DOI: 10.1038/s41467-017-00197-0 | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00197-0

ab Topography ACC front High EKE Isopycnal Trajectory Eddy transport

S. America

1000 m

1500 m Depth 2000 m

2500 m

Antarctica 30° S

Fig. 7 Idealized schematic illustrating the effect of eddy advection at topographic hotspots on upwelling pathways. a An idealized particle trajectory (red) follows time-mean Antarctic Circumpolar Current (ACC) streamlines (black) that flow southeastward around Antarctica from east of Drake Passage (blue surface indicating the particles’ isopycnal surface, lighter color indicating shallower depths). The trajectory crosses streamlines and upwells (yellow arrows) in regions of high eddy kinetic energy (EKE; yellow shading) at major topographic features (gray shading). This creates a superimposed southward/upward spiral as the particles shift southward and upward each time they encounter a region of high EKE. b A two-dimensional vertical cross-section of the Southern Ocean from Antarctica to 30° S, indicated by the white dashed line in a. White lines show idealized isopycnal layers shoaling and thinning toward the South. The red arrows show the trajectory entering the high EKE region associated with topography along the northernmost ACC front and exiting the region, shallower and further south (front positions indicated by dotted lines) variability52. The spatially varying upwelling identified here means models. The ocean component is based on the MOM5 code, and employs no mesoscale eddy parameterization in the tracer equation. A year 1990 control that Southern Ocean heat and carbon uptake estimates from sparse, fi fl simulation was used, with atmospheric CO2 xed at 355 p.p.m. CM2.6 is spun up ship-based observations are likely unreliable. New, year-round, oat- for 84 years preceding the period used for analysis. Twelve years of 5-day averaged based biogeochemical measurements are beginning to transform our velocity fields were used for the Lagrangian analysis. knowledge of the Southern Ocean carbon cycle, and will allow CESM is a high-resolution coupled climate model with nominal 1/10° ocean and 58 quantitative validation of the importance of topographic hotspots in sea-ice resolution and 1/4° atmosphere and land resolution . The ocean component uses the Parallel Ocean Program (POP2), with no mesoscale eddy parameterizations. the natural and anthropogenic carbon budgets. fi A year 2000 control simulation was used, with atmospheric CO2 xed at 367 p.p.m. Climate change is predicted to drive a strengthening in CESM is spun up for 80 years preceding the period used for analysis here. Twenty Southern Hemisphere westerly winds53, as has already been years of monthly averaged velocity fields were used for the Lagrangian analysis. observed in recent decades54. This trend has led to a more The SOSE is a 1/6°, data-assimilating, ocean general circulation model based on 55 the MIT General Circulation Model, configured in a domain from 24.7 to 78° S energetic eddy field in the ACC and is expected to drive a 59 56 with an open northern boundary and a sea ice model . No mesoscale eddy further increase in EKE in the ACC in the future . Our finding parameterization is employed. Using software developed by the consortium for that eddies play a key role in driving Southern Ocean upwelling Estimating the Climate and Circulation of the Ocean (http://www.ecco-group.org), indicates that upwelling rates are likely sensitive to wind-driven the SOSE assimilates the majority of available observations using an adjoint fi method. For this study we used the SOSE iteration 100 solution, which has been changes in the eddy eld. More vigorous eddies in the ACC could validated against ocean and ice observations41, and spans 6 years (2005–2010) with increase the supply of carbon-rich deep waters to the sea surface, 1-day averaged velocity fields used for the Lagrangian analysis. and hence may weaken the Southern Ocean carbon sink. In addition to comparisons of the global model ocean and atmospheric states with However, more work is needed to uncover the response of the observations, several papers specifically address the model representation of the ACC fi transport, Southern Ocean surface properties and overturning in CESM60, carbon sink to a change in the eddy eld. Similarly, changes in the 16, 61 41 fi CM2.6 , and SOSE . A comparison of the time-mean volume meridional eddy eld would likely also alter the supply of nutrients to the transport at 30° S in CESM, CM2.6, and SOSE shows reasonable agreement in the surface of the Southern Ocean, potentially altering the efficiency magnitude and spatial structure of volume transport (Fig. 2). The total southward of the biological pump. Our results demonstrate that a transport across 30° S between 1000 and 3500 m is 28.8, 22.7, and 32.9 Sv in the deep understanding of the three-dimensional upwelling in the CESM, CM2.6, and SOSE, respectively; the portion that does not upwell south of 30° S could be entrained into abyssal water without first reaching the sea surface, or cross Southern Ocean is needed to determine the complex role of the north of 30° S shallower than 1000 m. Estimated total southward transport from Southern Ocean in the global heat, carbon and nutrient budgets. hydrographic observations in this depth range is a comparable 18–30 Sv dependent on the choice of layer, which also include northward transport; in isopycnal layers, the maximized southward transport is order 42 Sv62.TheSouthernOceanupper Methods overturning cell has similar structure in the three models (Fig. 2), but the abyssal Observations. Mapping of hydrographic properties on neutral density surfaces overturning cell is significantly weaker in CM2.6. The transports were calculated on 27 was carried out using high-quality historical hydrographic data and the World potential density surfaces (referenced to 2000 m) online in CM2.6, using 30-day Ocean Circulation Experiment (WOCE) observations of the 1990s. The maps in averaged output in CESM and on neutral density surfaces using daily averaged output Figure 1a and Supplementary Figure 1 are derived from those in the WOCE in SOSE, which was remapped to approximate potential density surfaces41. 27 − Hydrographic Programme Southern Ocean Atlas , which used an objective The mixed layer depth in each model is calculated using an 0.03 kg m 3 density mapping technique with elliptical search radii, with longer spatial scales following threshold63. The upwelling pathways in all three models were found to be topographic contours. ACC fronts based on these hydrographic data are also insensitive to the mixed layer definition (not shown). Mean EKE at 1000 m in each 21 shown in Supplementary Figure 1 . model was calculated from the 1-day averaged velocities in SOSE, 5-day averaged velocities in CM2.6, and 30-day averaged velocities in CESM. In this case “eddies” fi fi Model simulations and state estimate.Offline Lagrangian analysis was per- are de ned as deviations from the long-term time-averaged velocity eld. formed in two global climate models (CM2.6 and CESM) and in the regional SOSE. CM2.6 is the high-resolution version of the Geophysical Fluid Dynamics Laboratory’s CM2-O coupled model suite57. It combines global nominal 1/10° Lagrangian methods. The same particle release experiment was conducted resolution ocean and sea ice models with 50 km resolution atmosphere and land offline with velocity output from each of the three models, using the

8 NATURE COMMUNICATIONS | (2017) 8:172 | DOI: 10.1038/s41467-017-00197-0 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00197-0 ARTICLE

Connectivity Modeling System64 (CMS) in CM2.6 and CESM and Octopus Received: 20 October 2016 Accepted: 9 June 2017 (http://github.com/jinbow/Octopus) in SOSE. In each case, >2.5 million particles were released at 30° S in every grid cell between 1000 and 3500 m depth. Particles were re-released at the same location every month for the duration of the model output velocities (6 years in SOSE, 12 years in CM2.6, and 20 years in CESM). The trajectories were integrated for a total of 200 years, looping through the model fi fi References output in time such that the velocity elds return to the rst time step once the end 1. Marshall, J. & Speer, K. Closure of the meridional overturning circulation of the output has been reached65. To avoid unphysical upwelling that might occur through Southern Ocean upwelling. Nat. Geosci. 5, 171–180 (2012). as a result of small model drifts when looping velocity output, the particle depths 2. Talley, L. D. Closure of the global overturning circulation through the Indian, are held constant during the looping time step. The time step for particle advection Pacific, and Southern Oceans: schematics and transports. Oceanography 26, was 1 h for the CMS experiments in CM2.6 and CESM, while for the Octopus – experiments in SOSE, the particle advection time step was 0.5 days. A 10-min 80 97 (2013). time step results in the same trajectories within a 100-day testing window because 3. Nihashi, S. & Ohshima, K. I. Circumpolar mapping of antarctic coastal (1) the SOSE velocities are saved as daily average and (2) a high order scheme polynyas and landfast sea ice: relationship and variability. J Clim. 28, – (fourth order Runge–Kutta) is used in the time integration. In Octopus, particles 3650 3670 (2015). are numerically reflected at the sea surface and water-land boundaries. In CMS, an 4. Lumpkin, R. & Speer, K. Global ocean meridional overturning. J. Phys. ad hoc boundary condition enforcing no-flux and no-slip boundary conditions is Oceanogr. 37, 2550–2562 (2007). imposed; however, 30% of released particles are lost to advection into topography 5. Toggweiler, J. & Samuels, B. On the ocean’s large-scale circulation near the limit within 200 years. It is unlikely that this loss significantly affected the upwelling of no vertical mixing. J. Phys. Oceanogr. 28, 1832–1852 (1998). pathways, as the particles lost to topography were strongly biased toward the 6. Wolfe, C. L. & Cessi, P. The adiabatic pole-to-pole overturning circulation. deepest particles with relatively low transport that were initially released near J. Phys. Oceanogr. 41, 1795–1810 (2011). topography at 30° S. However, it is possible that this difference in handling of 7. Watson, A. J. et al. Rapid cross-density ocean mixing at mid-depths in the particles at the boundary could have contributed to the relatively large upwelling drake passage measured by tracer release. Nature 501, 408–411 (2013). particle transport in SOSE, where no particles are lost at the boundaries. 8. Nikurashin, M. & Ferrari, R. Global energy conversion rate from geostrophic flows There is no parameterization of small scale mixing used in the Lagrangian into internal lee waves in the deep ocean. Geophys. Res. Lett. 38, L08610 (2011). experiments, but a comparison in SOSE shows that upwelling pathways are 9. Naveira Garabato, A. C., Polzin, K. L., Ferrari, R., Zika, J. D. & Forryan, A. relatively insensitive to the inclusion of a stochastic noise component to represent A microscale view of mixing and overturning across the Antarctic Circumpolar sub-grid scale diffusion (Supplementary Fig. 4). Current. J. Phys. Oceanogr. 46, 233–254 (2016). After 200 years of particle advection, only particles that reached the surface 10. Ferrari, R., Mashayek, A., McDougall, T. J., Nikurashin, M. & Campin, J.-M. mixed layer and remained south of 30° S were selected for analysis. Less than 5% of Turning ocean mixing upside down. J. Phys. Oceanogr. 46, 2239–2261 (2016). fi the total released particle trajectories ful lled these criteria in all three simulations, 11. Wåhlin, A. K., Yuan, X., Björk, G. & Nohr, C. Inflow of warm circumpolar deep leaving ~100,000 trajectories in each. Of the remaining 95% of particles released water in the Central Amundsen shelf. J. Phys. Oceanogr. 40, 1427–1434 (2010). that did not upwell, approximately half of the particles are excluded because 12. Rignot, E., Jacobs, S., Mouginot, J. & Scheuchl, B. Ice-shelf melting around they had initial northward velocities and the majority of the remainder exit north Antarctica. Science 341, 266–270 (2013). of 30° S without upwelling, leaving <1.5% of particles south of 30° S that did not 13. Paolo, F. S., Fricker, H. A. & Padman, L. Volume loss from Antarctic ice shelves upwell into the mixed layer during the 200 year experiment. We only considered is accelerating. Science 348, 327–332 (2015). the portions of trajectories before particles reach the mixed layer. Pathways in Fig. 2 are insensitive to different mixed layer depth definitions or using a constant 14. Hallberg, R. & Gnanadesikan, A. The role of eddies in determining the depth crossing of 200 m rather than mixed layer depth (data not shown). structure and response of the wind-driven southern hemisphere overturning: It is common to use Lagrangian particle tracking to assess volume transports results from the modeling eddies in the southern ocean (MESO) project. J. Phys. – between chosen sites in the ocean, by assigning a volume transport to each particle Oceanogr. 36, 2232 2252 (2006). – at its release31, 66 69. Each particle was “tagged” with the meridional volume 15. Morrison, A. K. & Hogg, A. M. On the relationship between Southern Ocean transport (in Sverdrups) at its release location at 30° S by multiplying the overturning and ACC transport. J. Phys. Oceanogr. 43, 140–148 (2013). meridional velocity by the area of the model grid cell at the particle release location. 16. Dufour, C. et al. Role of mesoscale eddies in cross-frontal transport of heat and Because the model velocity fields are non-divergent, and a sufficient number of biogeochemical tracers in the Southern Ocean. J. Phys. Oceanogr. 45, particles are released to allow for deformation of the flow, it is assumed that 3057–3081 (2015). the transport carried by each particle is conserved over the length of the 17. Toggweiler, J. R. & Samuels, B. Effect of drake passage on the global simulation31, 66. This volume transport is then conserved along the trajectories thermohaline circulation. Deep Sea Res. Pt. I 42, 477–500 (1995). until they reach the mixed layer, providing an estimate of the transport of 18. Sallée, J. B., Speer, K. G. & Rintoul, S. R. Zonally asymmetric response of the upwelling deep water between 30° S and the mixed layer. southern ocean mixed-layer depth to the southern annular mode. Nat. Geosci. Particle transport weighting was used in Figs. 1 and 4–6, by summing the 3, 273–279 (2010). volume transports of each particle at each location, and normalizing by the total 19. Thompson, A. F. & Sallee, J.-B. Jets and topography: jet transitions and the volume transport of all of the particles. Therefore, particles assigned with more impact on transport in the Antarctic Circumpolar Current. J. Phys. Oceanogr. transport initially have a larger contribution to the pathway distributions. Our 42, 956–973 (2012). spatial upwelling pathways are qualitatively unaffected by this transport weighting, 20. Naveira Garabato, A. C., Ferrari, R. & Polzin, K. L. Eddy stirring in the although it does affect the relative timescales and strengths of different pathways. Southern Ocean. J. Geophys. Res. 116, C09019 (2011). We refer to particle trajectories weighted by their initial transport at 30° S as 21. Orsi, A. H., Whitworth, T. & Nowlin, W. D. On the meridional extent and fronts “particle transport”. The accuracy of the resulting transport pathways depends on – fl of the Antarctic Circumpolar Current. Deep Sea Res. Pt. I 42, 641 673 (1995). the number of particles deployed and complexity of the ow, so the accuracy of 22. Döös, K. Interocean exchange of water masses. J. Geophys. Res. 100, 13499 particle-transport pathways was tested by randomly halving the selection of (1995). particles and was found to be insensitive (Supplementary Fig. 5). Other recent 23. WAIS Divide Project Members. Precise interpolar phasing of abrupt climate experiments in eddy-resolving models show good agreement between Lagrangian change during the last ice age. Nature 520, 661–665 (2015). transports and Eulerian transports on decadal timescales68. 24. Frölicher, T. L. et al. Dominance of the Southern Ocean in anthropogenic carbon and heat uptake in CMIP5 models. J. Clim. 28, 862–886 (2015). Code availability. SOSE is based on the MITgcm code framework, available 25. Sarmiento, J. L., Gruber, N., Brzezinski, M. A. & Dunne, J. P. High-latitude at http://mitgcm.org. Code to run the CM2.6 experiment is available from http:// controls of thermocline nutrients and low latitude biological productivity. www.gfdl.noaa.gov/cm2-5-and-flor. The CMS, used for Lagrangian experiments with Nature 427,56–60 (2004). CESM and CM2.6, is an open-source Fortran toolbox available for download at 26. Chelton, D. B., Schlax, M. G., Witter, D. L. & Richman, J. G. Geosat altimeter https://github.com/beatrixparis/connectivity-modeling-system. The Octopus Lagran- observations of the surface circulation of the Southern Ocean. J. Geophys. Res. gian code, used with SOSE, is available at https://github.com/jinbow/Octopus.Ana- 95, 17877–17903 (1990). fi lysis was completed using the open-source Python scienti c stack (http://scipy.org). 27. Orsi, A. H. & Whitworth, T. Hydrographic Atlas of the World Ocean Circulation Experiment (WOCE): Volume 1: Southern Ocean (WOCE International Project fi Data availability. All data supporting this study are open and freely available. Of ce, 2005). Hydrographic section data from the WOCE atlas are available at the NOAA 28. Cook, A. J. et al. Ocean forcing of glacier retreat in the western Antarctic National Centers for Environmental Information (https://www.ncei.noaa.gov/) and Peninsula. Science 353, 283–286 (2016). the CLIVAR and Carbon Hydrographic Data Office (http://cchdo.ucsd.edu). 29. Iudicone, D., Speich, S., Madec, G. & Blanke, B. The global conveyor belt from a The model output from the CESM model is available through the Earth System Southern Ocean perspective. J. Phys. Oceanogr. 38, 1401–1425 (2008). Grid (http://earthsystemgrid.org). Output from CM2.6 used to generate figures in 30. Van Sebille, E., Johns, W. E. & Beal, L. M. Does the vorticity flux from Agulhas this paper are available from the corresponding author on reasonable request. rings control the zonal pathway of NADW across the South Atlantic? SOSE Iteration 100 model output is available at http://sose.ucsd.edu. J. Geophys. Res. 117, C05037 (2012).

NATURE COMMUNICATIONS | (2017) 8:172 | DOI: 10.1038/s41467-017-00197-0 | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00197-0

31. Döös, K., Nycander, J. & Coward, A. C. Lagrangian decomposition of the 64. Paris, C. B., Helgers, J., van Sebille, E. & Srinivasan, A. Connectivity modeling deacon cell. J. Geophys. Res. 113, C07028 (2008). system: a probabilistic modeling tool for the multi-scale tracking of biotic and 32. McDonagh, E. L., Bryden, H. L., King, B. A. & Sanders, R. J. The circulation of abiotic variability in the ocean. Environ. Model. Softw. 42,47–54 (2013). the Indian Ocean at 32 °S. Prog. Oceanogr. 79,20–36 (2008). 65. Van Sebille, E. et al. Abyssal connections of Antarctic bottom water in a 33. Reid, J. L. On the total geostrophic circulation of the North-Atlantic Ocean: Southern Ocean state estimate. Geophys. Res. Lett. 40, 2177–2182 (2013). flow patterns, tracers, and transports. Prog. Oceanogr. 33,1–92 (1994). 66. Blanke, B., Arhan, M., Madec, G. & Roche, S. Warm water paths in the 34. Garzoli, S. L. et al. The fate of the deep western boundary current in the South equatorial Atlantic as diagnosed with a general circulation model. J. Phys. Atlantic. Deep Sea Res. Pt. I 103, 125–136 (2015). Oceanogr. 29, 2753–2768 (1999). 35. Hufford, G. E., McCartney, M. S. & Donohue, K. A. Northern boundary 67. Speich, S., Blanke, B. & Madec, G. Warm and cold water routes of an ogcm currents and adjacent recirculations off southwestern australia. Geophys. Res. thermohaline conveyor belt. Geophys. Res. Lett. 28, 311–314 (2001). Lett. 24, 2797–2800 (1997). 68.VanSebille,E.etal.Pacific-to-indian ocean connectivity: tasman leakage, indonesian 36. Talley, L. D. in Hydrographic Atlas of the World Ocean Circulation Experiment throughflow, and the role of enso. J. Geophys. Res. 119, 1365–1382 (2014). (WOCE), Vol. 4 (eds Sparrow, M., Chapman, P. & Gould, J.) Indian Ocean 69. Thomas, M. D., Tréguier, A.-M., Blanke, B., Deshayes, J. & Voldoire, A. A (International WOCE Project Office, 2013). lagrangian method to isolate the impacts of mixed layer subduction on the 37. Speich, S. et al. Tasman leakage: a new route in the global ocean conveyor belt. meridional overturning circulation in a numerical model. J. Clim. 28, Geophys. Res. Lett. 29, 1416 (2002). 7503–7517 (2015). 38. Reid, J. L. On the total geostrophic circulation of the Pacific Ocean: flow 70. Smith, W. H. & Sandwell, D. T. Global sea floor topography from satellite patterns, tracers, and transports. Prog. Oceanogr. 39, 263–352 (1997). altimetry and ship depth soundings. Science 277, 1956–1962 (1997). 39. Faure, V. & Speer, K. Deep circulation in the eastern south pacific ocean. 71. Ramachandran, P. & Varoquaux, G. Mayavi: 3D visualisation of scientific data. J. Marine Res. 70, 748–778 (2012). Comput. Sci. Eng. 13,40–51 (2011). 40. Witter, D. L. & Chelton, D. B. Eddy-mean flow interaction in zonal oceanic jet fl – ow along zonal ridge topography. J. Phys. Oceanogr. 28, 2019 2039 (1998). Acknowledgements 41. Abernathey, R. P. et al. Water-mass transformation by sea ice in the upper branch of the Southern Ocean overturning. Nat. Geosci. 9, 596–601 (2016). V.T., L.D.T., and M.R.M. were supported by NSF OCE-1357072. A.K.M., H.F.D., and W.W. were supported by the RGCM program of the US Department of Energy under 42. Viglione, G. A. & Thompson, A. F. Lagrangian pathways of upwelling in the ’ Southern Ocean. J. Geophys. Res. 121, 6295–6309 (2016). Contract DE-SC0012457. J.L.S. acknowledges NSF s Southern Ocean Carbon and 43. Sloyan, B. M. & Rintoul, S. R. Circulation, renewal, and modification of antarctic Climate Observations and Modeling project under NSF PLR-1425989, which partially mode and intermediate water*. J. Phys. Oceanogr. 31, 1005–1030 (2001). supported L.D.T. and M.R.M. as well. C.O.D was supported by the National Aeronautics 44. Garabato, A. C. N., Polzin, K. L., King, B. A., Heywood, K. J. & Visbeck, M. and Space Administration (NASA) under Award NNX14AL40G and by the Princeton Widespread intense turbulent mixing in the southern ocean. Science 303, Environmental Institute Grand Challenge initiative. A.R.G. was supported by a Climate 210–213 (2004). and Global Change Postdoctoral Fellowship from the National Oceanic and Atmospheric 45. Katsumata, K., Sloyan, B. & Masuda, S. Diapycnal and isopycnal transports in Administration (NOAA). S.M.G. acknowledges the ongoing support of NOAA/GFDL for the southern ocean estimated by a box inverse model. J. Phys. Oceanogr. 43, high-end ocean and climate-modeling activities. J.W. acknowledges support from – NSF OCE-1234473 and declare that this work was done as a private venture and not in 2270 2287 (2013). ’ 46. Sheen, K. et al. Eddy-induced variability in southern ocean abyssal mixing on the author s capacity as an employee of the Jet Propulsion Laboratory, California climatic timescales. Nat. Geosci. 7, 577–582 (2014). Institute of Technology. Computational resources for the SOSE were provided by NSF 47. Rhein, M., Kieke, D. & Steinfeldt, R. Advection of north atlantic deep water XSEDE resource grant OCE130007. Output from the ASD run of CESM was made available from the labrador sea to the southern hemisphere. J. Geophys. Res. 120, by J. Small (NCAR) and colleagues. We thank Alejandro Orsi for the gridded hydrographic 2471–2487 (2015). observations, Stu Bishop for providing processed CESM output, and Ryan Abernathey, 48. Armour, K. C., Marshall, J., Scott, J. R., Donohoe, A. & Newsom, E. R. Southern Haidi Chen, and Nathaniel Tarshish for useful discussions on the manuscript. Ocean warming delayed by circumpolar upwelling and equatorward transport. Nat. Geosci. 9, 549–554 (2016). Author contributions 49. Le Quéré, C. et al. Saturation of the Southern Ocean CO2 sink due to recent V.T. and H.F.D. performed Lagrangian experiments in SOSE, CM2.6, and CESM; J.W. wrote climate change. Science 316, 1735–1738 (2007). the Octopus Lagrangian code for SOSE; V.T., H.F.D., and A.K.M. analyzed model output and 50. Lovenduski, N. S., Gruber, N. & Doney, S. C. Toward a mechanistic Lagrangian trajectories; L.D.T. analyzed the hydrographic data that motivated the model understanding of the decadal trends in the Southern Ocean carbon sink. Global study; all authors contributed to interpretation of the results and writing of the manuscript. Biogeochem. Cycles 22, GB3016 (2008). 51. Ito, T., Woloszyn, M. & Mazloff, M. Anthropogenic carbon dioxide transport in the Southern Ocean driven by Ekman flow. Nature 463,80–83 (2010). Additional information 52. Landschützer, P. et al. The reinvigoration of the Southern Ocean carbon sink. Supplementary Information accompanies this paper at https://doi.org/10.1038/s41467- Science 349, 1221–1224 (2015). 017-00197-0. 53. Swart, N. C. & Fyfe, J. C. Observed and simulated changes in the Southern fi hemisphere surface westerly wind-stress. Geophys. Res. Lett. 39,6–11 (2012). Competing interests: The authors declare no competing nancial interests. 54. Marshall, G. Trends in the southern annular mode from observations and reanalyses. J. Clim. 16, 4134–4143 (2003). Reprints and permission information is available online at http://npg.nature.com/ 55. Hogg, A. M. et al. Recent trends in the Southern Ocean eddy field. J. Geophys. reprintsandpermissions/ Res. 120, 257–267 (2015). 56. Patara, L., Böning, C. W. & Biastoch, A. Variability and trends in Southern Change history: A correction to this article has been published and is linked from the Ocean eddy activity in 1/12° ocean model simulations. Geophys. Res. Lett. 43, HTML version of this paper. 4517–4523 (2016). 57. Griffies, S. M. et al. Impacts on ocean heat from transient mesoscale eddies in a Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in fi hierarchy of climate models. J. Clim. 28, 952–977 (2015). published maps and institutional af liations. 58. Small, R. J. et al. A new synoptic scale resolving global climate simulation using the community earth system model. J. Adv. Model. Earth Syst. 6, 1065–1094 (2014). Open Access This article is licensed under a Creative Commons 59. Mazloff, M. R., Heimbach, P. & Wunsch, C. An eddy-permitting southern Attribution 4.0 International License, which permits use, sharing, ocean state estimate. J. Phys. Oceanogr. 40, 880–899 (2010). adaptation, distribution and reproduction in any medium or format, as long as you give 60. Bishop, S. P. et al. Southern ocean overturning compensation in an eddy- appropriate credit to the original author(s) and the source, provide a link to the Creative resolving climate simulation. J. Phys. Oceanogr. 46, 1575–1592 (2016). Commons license, and indicate if changes were made. The images or other third party 61. Morrison, A. K., Griffies, S. M., Winton, M., Anderson, W. G. & Sarmiento, J. L. material in this article are included in the article’s Creative Commons license, unless Mechanisms of Southern Ocean heat uptake and transport in a global eddying indicated otherwise in a credit line to the material. If material is not included in the climate model. J. Clim. 29, 2059–2075 (2016). article’s Creative Commons license and your intended use is not permitted by statutory 62. Talley, L. D. Freshwater transport estimates and the global overturning regulation or exceeds the permitted use, you will need to obtain permission directly from circulation: shallow, deep and throughflow components. Prog. Oceanogr. 78, the copyright holder. To view a copy of this license, visit http://creativecommons.org/ 257–303 (2008). licenses/by/4.0/. 63. de Boyer Montégut, C., Madec, G., Fischer, A. S., Lazar, A. & Iudicone, D. Mixed layer depth over the global ocean: an examination of profile data and a profile-based climatology. J. Geophys. Res. 109, C12003 (2004). © The Author(s) 2018

10 NATURE COMMUNICATIONS | (2017) 8:172 | DOI: 10.1038/s41467-017-00197-0 | www.nature.com/naturecommunications